Critical Appraisal: Epidemiology 101
POS Lecture Series April 28, 2004
What to Believe?
"A proof is a proof. What kind of a proof? It's a proof. A proof is a proof. And when you have a good proof, it's because it's proven."
Introduction
Why do I need Critical Appraisal Skills?
– Not all literature accurate – Conclusions drawn not always possible – Why the inaccuracies?
• Stupidity • “Publish or perish” • Money
– Being cynical and suspicious is healthy
The best defense is to be prepared
Introduction
• Types of studies • Important components of a good randomized trial • 6 important questions to ask yourself when reading a paper
Study Types
• Descriptive, Observational, Experimental
– Descriptive – series, case report – Observational – groups determined by predetermined factor – Experimental – investigator in control of group assignments
Types of Studies Observational
• Case-control
– uses – Advantages and disadvantages
• Cost, good for causation in rare disease • Recall bias
Types of Studies Observational
• Cohort
– Definition
• Advantages and disadvantages • Prospective • Cost high
– Esp if disease is rare or time between exposure and onset of disease is long
Types of Studies Experimental
• Randomized trial • “Gold Standard”
– Advantages and disadvantages
Principles of a Good Trial
• Ideas, research question, hypothesis
– Clinical relevance – Is it possible?
• Time, finances, ethics
Principles of a Good Trial
• Literature search
– Background – Results of other trials – Convinced it was extensive
Principles of a Good Trial
• Patient Selection
– Inclusion and exclusion criteria
• • • • Are they well defined? Are they reasonable? Are they clinically relevant? Do they change the results?
Principles of a Good Trial
• Sample size calculation
– Most ortho literature does not mention – There is SOME science – Based on primary outcome measurement
Sample Size Calculation
• n = 2 [( + ) / ] 2 • Z of α (Type one error)
– Usually 0.05 z=1.96
• Z of β (Type II error)
– Usually 0.2 Z=1.28
Sample Size Calculation
• n = 2 [( + ) / ] 2 • = S.D. of outcome measure
– How do you know??
• Pilot study • published
Sample Size Calculation
• n = 2 [( + ) / ] 2 • = Clinically relevant difference
– This is the variable that can be manipulated – Depends of risks/cost of treatment
Sample Size Calculation
• n = 2 [( + ) / ] 2 • Equivalency trial
– Rarely done =0.05 and sample size increases
• A neg trial that does not address this can not conclude “no difference in treatments” only “we failed to prove a difference”
Randomization
• Computer, random number table, coin toss • Not birthday, MCP • Block randomization
– Small number, multi-center – AABB, ABBA, etc – Potential for bias
Blinding
• Always adds weight to a study
– Are the subject and investigators blinded – Is it feasable or possible?
Intervention
• Well defined, particulars discussed
Outcome Measurement
• Primary outcome measure • Secondary outcome measures
– Data dredging
Analysis
• Biostats
– Definitely some trust here – Everyone can’t be an expert
Relative Risk Reduction (RRR)
Unreamed Non-Union Rate .1 Reamed .05
RRR = (0.1 – 0.05)/ 0.1 = 50% If outcome is rare, this is misleading
Absolute Risk Reduction (ARR)
Unreamed Non-Union Rate .1 Reamed .05
ARR = 0.1 – 0.05 = 5% Good for rare outcomes and NNT
Number Needed to Treat (NNT)
Unreamed Non-Union Rate .1 Reamed .05
ARR = 0.1 – 0.05 = 5% NNT = 1/ARR = 1/0.05 = 20
Lost to Follow-up
• 20 % added to sample size • Good Investigators very aggressive • “Worse case” Analysis
Six Questions to Ask before you change your practice!
1. Really Randomized?
2. All clinically relevant outcomes Reported?
3. Patients similar to your own?
4. Was clinical and statistical significant considered?
5. Is the intervention feasible in your practice?
6. All patients accounted for?